Description
The deflection of a bridge is a crucial factor in assessing its structural integrity and safety, serving as an indicator of the overall rigidity of the bridge. The deflections stemming from temperature variations may surpass those attributed to live loads. However, the monitoring data acquired from the sensors indicate a time-lag effect exists between temperature variations and resulting deflection. The time-lag effect poses challenges in precisely characterizing and modeling the temperature-induced deflection behavior. Therefore, this paper presents a deep learning-based prediction model to forecast the temperature-induced deflection of strengthened multi-span continuous box girder bridges. The Long-Short Term Memory (LSTM) model was adapted in this paper leveraging deep learning techniques which capable to learn complex patterns and non-linear relationships between temperature and temperature-induced deflection data to forecast the deflection under different temperature variations. To enhance the precision of the LSTM model, this paper proposed the Empirical Mode Decomposition (EMD) method to extract the temperature-induced deflection from the raw deflection data which induced by other various factors such as vehicles loads and wind loads. In the forecasting outputs obtained from the LSTM model, the root mean squared error (RMSE) and mean absolute error (MAE) were noteworthy achieved a low error value of 0.51mm and 0.39mm respectively. Lastly, residual analysis was conducted to analyze the forecast outputs from the LSTM model with the actual measurements obtained from sensors.
